Benchmark Almanac

Winogender

Gemma 2 9B 79.2% leads since July 2024
Only in system cards

Models resolve pronouns in sentences about occupations. Accuracy shows whether gender stereotypes sway coreference.

Released
April 2018
Kind
Bias, fairness and politics
Level
Everyday
Scored
Checked automatically
Answers
Multiple choice
Tasks
All public
Strongest model tested
Gemma 2 27B #189 on the capability index
Last in a system card
July 2024
Last model added
July 2024

In the authors’ words

We present an empirical study of gender bias in coreference resolution systems. We first introduce a novel, Winograd schema-style set of minimal pair sentences that differ only by pronoun gender. With these "Winogender schemas," we evaluate and confirm systematic gender bias in three publicly-available coreference resolution systems, and correlate this bias with real-world and textual gender statistics.

Abstract of Gender Bias in Coreference Resolution (arXiv:1804.09301, CC0)
4
Models scored in lab cards
4
System cards report it, from 1 lab

Gemma 2 9B leads with 79.2% among models released by June 2024

Best score per model as labs report them in their own cards. Labs pick their own settings, so these compare less cleanly than independent runs.

The newest model with a score here came out in June 2024, and nobody has reported a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 79.2 Gemma 2 9B
  2. 77.2 Gemma 2 27B
  3. 54.2 Gemma 7B
  4. 51.3 Gemma 2B
  1. 1 Gemma 2 9B 79.2%
  2. 2 Gemma 2 27B 77.2%
  3. 3 Gemma 7B 54.2%
  4. 4 Gemma 2B 51.3%

Source: Labs’ model cards

#ModelScoreSettingDateSource
1 Gemma 2 9B 79.2% Jul 2024 Lab model card
2 Gemma 2 27B 77.2% Jul 2024 Lab model card
3 Gemma 7B 54.2% Mar 2024 Lab model card
4 Gemma 2B 51.3% Mar 2024 Lab model card

2 models have held number one in labs’ own numbers since 2023

Lab claims count from the day the card came out, independent runs from the model’s release.

Google DeepMind
Lab claims Not tested since Jul 2024
Gemma 2 9B 79.2%

Source: Labs’ model cards

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

Reported by the lab in its model card Best so far
0% 20% 40% 60% 80% 100% 2024 Near the ceiling (85%) Gemma 2 9B Gemma 7B 0% 20% 40% 60% 80% 100% 2024 Near the ceiling (85%) Gemma 2 9B Gemma 7B

Source: Labs’ model cards

Labs report it only in system cards

We read the headline results of 522 launch posts from 52 labs, and none of them list this benchmark. Some system cards do, below.

4 system cards report it, from 1 lab

Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.

DateDocumentScoreAs the document writes itSection
Jul 31, 2024 Gemma 2 9B technical report 79.2 Winogender, top-1, IT (Table 18) Bias
Jul 31, 2024 Gemma 2 27B technical report 77.2 Winogender, top-1, IT (Table 18) Bias
Mar 13, 2024 Gemma 7B technical report 54.2 Winogender, top-1, IT (Table 8) Bias
Mar 13, 2024 Gemma 2B technical report 51.3 Winogender, top-1, IT (Table 8) Bias

More in Safety and alignment

All 465 in Safety and alignment
  • FORTRESS In use Paired adversarial and benign national-security misuse prompts. Rubric-graded risk score on adversarial and over-refusal score on benign prompts. Claude Opus 5.5 5.78%
  • StrongREJECT In use Forbidden-prompt jailbreak benchmark with a rubric grader that scores how useful a non-refusal actually is. Inkling 98.6%
  • Gray Swan IPI Arena benchmark In use Indirect prompt-injection attacks from public red-teaming competitions replayed on tool, coding and browser agents. Attack success rate at k attempts. Gemini 4 Carbon 0%
  • DistressBench In use Models reply as a chatbot to 718 simulated users in psychological crisis, in single-turn and multi-turn chats. Three judge models score replies against clinician-written rubrics. Muse Spark 1.2 88.4%
  • Lakera B3 (Backbone Breaker Benchmark) In use Crowdsourced prompt-injection attacks hit backbone LLMs inside agent threat snapshots at three defense levels. The score is attack resistance, 100 minus the board's vulnerability percentage. Claude Opus 4.5 89%
  • IHEval In use Models face conflicts between system, user and tool instructions. Reports a composite score for following the higher-priority instruction. GPT-4o 70%